|-- drive_manually.py (with flag --collect_data, stores manual driving data in './data/')
|-- model.py (agent in './models/')
|-- train_agent.py (train agent)
|-- test_agent.py (agent performance, stores results in './results/')
Install pyTorch and gym requirements (I used an anaconda Python3.6 virtual env).
Neural Network has a validation MSE loss of 0.06 and mean episode rewards of 450 using a current image input grayscaled to (1,96,96) and preprocessed.
python test_agent.py
gives the following results:
